20VC：Anthropic获Microsoft与NVIDIA合计300亿美元支持｜TPU威胁NVIDIA核心业务｜解析Sam Altman的“战争模式”｜Sierra年经常性收入达1亿美元：能否支撑100亿美元估值？｜Lovable年经常性收入达2亿美元，传将完成60亿美元融资

Chamath Palihapitiya

But before we dive into the show today... Now, most people who get scammed never talk about it, and if it can happen to tech-savvy professionals, CEOs, and investors, it can happen to anyone. But the problem isn't just losing money, it's that today's scams, they're built differently for a very new world, one where AI can generate convincing messages in seconds, and fake sites look more like real sites than the real thing. Traditional tools were not built for this future, and that's why Guardio exists. Guardio is this incredible predictive and proactive engine. It leverages advanced AI threat detection to block highly targeted, socially engineered scams before they ever reach you, from phishing emails and fake login pages to financial fraud. Guardio protects you across the ways people actually live and work online. And security shouldn't be complicated. Guardio continuously monitors across all your accounts and devices, uncovering risks in real time and guiding you to close gaps before attackers exploit them. Trusted by over a million users, Guardio is setting the new standard for personal cybersecurity. Visit guard.io/20vc today to start your seven-day free trial. Because the threats of tomorrow, they're already here, and Guardio is built to stop them. After Guardio protects you from what's out there on the internet, Squarespace helps you put your work out there. Every dream needs a domain. Bring it to life with Squarespace. Your custom domain comes with everything it should: powerful built-in security, award-winning support, and tools that help you grow for an upfront, transparent fee. No matter where you are in your journey, do more with a domain. Go to domains.squarespace.com/20vc and get 20% off your dream domain with coupon code VC20. That's VC20. And finally, we have to speak about our newest sponsor. It's Intercom. If you're looking for a way to transform your customer service, let me introduce you to Fin, baby. Fin is the number one AI agent for customer service, resolving up to 93% of customer queries automatically. There is no other agent that can do that. Not 93% of customer queries, okay? No other agent can do that. So why choose Fin? Fin is the best performing AI agent for CS. Fin doesn't just answer questions. It takes actions. It automates the most complex customer queries like refunds, transaction disputes, technical troubleshooting with speed and reliability. I wish my team was speedy and reliable. Beats every competitor in every head-to-head bake-off. Completely configurable and code optional setup. My word, I mean, the benefits just go on and on. It's easy and efficient implementation. It works on any help desk with no tedious migration needs. It's trusted by over 6,000 customer service leaders, including top AI companies like Anthropic, Lovable, Synthesia, Clay, Vanta. So if you're ready to transform your customer service team, scale your support, and give team members time to focus on the really high-level strategic work, learn more about Fin at fin.ai/20vc.

Rory O'Driscoll

You have now arrived at your destination.

Rory O'Driscoll

The P/E on NVIDIA, to give that soundbite that I love, is lower than the P/E on Costco.

Jason Calacanis

I don’t care about your talk. I don’t want to hear about your pilot. I want to smell that your team is in hyper-aggressive mode. If it is, I think you can come back. I do not think you can push your team too hard. I think you should push them as hard as the business needs to go, and if they leave, it’s great.

Salesforce could be the next Google. In 18 to 24 months, we could be like, “Oh my God, Agentforce crushed it. All these startups, we didn’t need them because Agentforce is so good.”

Chamath Palihapitiya

The great American tech Magnificent 7 suck more of the world’s profit dollars out of the rest of humanity. Go team.

Jason Calacanis

It is very exciting to be at the new upgraded 20VC Studios. It’s exciting here.

I mean, thank you. Rory, we miss you. It’s not the same without you here, but thank you for joining us.

Chamath Palihapitiya

You’re welcome. You’ll survive.

1. Anthropic Raises $30 Billion

You know what? We will, but we want to start with the news of the day. Anthropic secured up to $15 billion from Microsoft and NVIDIA, pushing its valuation to $350 billion, with commitments for $30 billion in Azure compute. You can see I’m learning from prior episodes in setting the context. Let’s start there. How did we analyze that?

Jason Calacanis

It was pretty funny watching Twitter this week. A week ago, they were like, “OpenAI is dominant.” Three days ago, it was like, “Gemini 3 Pro is...” Mark Benioff was like, “I’ll never go back to ChatGPT again,” and today it’s, “Claude 4.5 has killed, crushed everything.”

My point is, there’s just no stability. There’s no stability in seed investing. There’s no stability in Anthropic. So my meta-learning, which I wouldn’t have even had a couple of weeks ago, is more power to them, because I think you need infinite capital when there’s no stability.

It’s just so funny. Literally 3 days ago, it was Gemini, and then today it’s Anthropic. What’ll it be next week?

Chamath Palihapitiya

But there was a bunch of stuff in the announcement. First of all, you glossed over the key fact: it was from Microsoft.

Mm-hmm.

Chamath Palihapitiya

This was the NVIDIA–Microsoft commitment to Anthropic, and Microsoft was in a monogamous relationship with OpenAI. Then OpenAI wanted an open marriage, and Microsoft said, “Well, if you want an open marriage, I want one too,” right?

Jason Calacanis

Yep.

Chamath Palihapitiya

So this was probably inevitable in that context. That’s probably the first big piece of news from it: the Microsoft and NVIDIA commitment.

Then there’s the structure of the deal. It’s the usual thing: you get $15 billion, and you promise to spend $30 billion. We’ve seen that structure before.

The other interesting thing—you didn’t mention it—is that in the same announcement, I think Anthropic also said they’re going to break ground on a physical data center. So that’s another one of the model companies saying it’s not enough to rely on compute from your service providers, whether it’s Azure or AWS. They’re also looking at doing physical data centers themselves.

So it’s another chapter in the infinite-capital-and-everybody’s-sleeping-with-everybody war.

Jason Calacanis

Have we just decided we don’t care about round-trip revenue? Like many things in this era, we’ve just given up caring. There are so many things we used to care about in the past, and now we just want to get rich with AI. We don’t care.

Chamath Palihapitiya

In a bull market, nobody cares about anything, and then in a bear market, everybody discovers why you were meant to care. Now we’re in the don’t-care part of the trade.

Jason Calacanis

Don’t care.

Chamath Palihapitiya

Right? And again, as we said before, provided it works, it’s all fine, right? For what it’s worth, if you’re a chip provider and you had to stick some money in one model provider, this one feels like a pretty good bet.

So, zooming out, of all the round-tripping deals, I would argue a Microsoft–Anthropic–NVIDIA deal probably has better principles and upside than most. Some of the other round-tripping deals look like they’re already a bit shaky.

2. The TPU Verticalization Race

David Friedberg

Totally get you there. You said something about the importance of verticalization in terms of the data-center play that’s been added onto that deal. Another very important bit of news was that Google trained Gemini 3. Obviously, people were very impressed with the quality of Gemini 3, Benioff included, saying that it’s trained on its own TPUs—their own chips, for people who aren’t aware of TPUs.

Elon’s AI company, xAI, is developing its own AI inference chips. On the question of verticalization, as we said there, with owning the data-center layer, does everyone now need to own the chip layer as well, as we see more and more with TPUs and with Elon? Is that the next phase of verticalization?

Jason Calacanis

As you guys know, I use Replit 2 hours a day, a little less on my London trip. But it is good to learn, right? I’m in the top 1% of users.

Lovable’s great too. They added Gemini 3 Pro the day it came out, as did everybody. I instantly used it. It was great. It actually wasn’t so much better. What everyone said is that it’s better for design. Lovable said it, Replit, everyone said it. I would say it was 20% better for design, but that wasn’t the interesting part.

I used it, it worked great, and then in my next prompt, I rolled back to Claude. It was fine. So not only are there different models, but abstracting a couple of layers above, we’re switching between TPUs and models, and I don’t care. I’m not saying these aren’t huge issues, but the idea that NVIDIA is unstoppable because of the software-and-hardware connection, because we have to have GPUs—I know there’s a lot of truth to that, but literally, as an end user, I went right back and forth with no issue.

TPUs, GPUs, and LLMs—all in the space of 60 seconds, or maybe 3 minutes.

Chamath Palihapitiya

The big-picture question you were asking is, does every large, vertically integrated company have to do its own chips versus buying from NVIDIA? The reason you pose that question is that Google obviously has its own TPUs, which are internal chips that I believe are faster for what they do.

The interesting thing here is that you’ve got to separate 90% by customer count of NVIDIA’s customers. If you’re spending $1 million, $2 million, $5 million, or $10 million with NVIDIA, it’s in the noise, and you’re not going to design and build your own chip. That would be madness.

The odd thing about the NVIDIA business, unlike most other businesses, is that 4 or 5 of its customers account for 80% of the revenue—something like that, 70% to 80%. Looking at the other side, if you’re spending...

Jason Calacanis

Google is spending $90 billion this year on CapEx. A rough rule of thumb is that around 40% of the total dollars are on compute. So if they were buying NVIDIA chips, they'd be spending $36 billion on compute, literally just on chips. If you're spending $36 billion, and NVIDIA has 75% plus gross margins, which means you're handing NVIDIA, if you were buying all that from NVIDIA, north of $20 billion a year of profit at the margin.

At that point, you say to yourself, “Well, it's hard to build a chip, and if I was spending $10 million, I wouldn't bother. It's probably gonna cost me…” Back when venture guys did chips, it was $200 million to get to a chip. Probably today it's $1 billion. But if you're giving someone $20 billion of profit a year, and you can say to yourself, “Maybe I can invest $1 billion a year for 5 years, get a compelling chip,” you gotta look at that.

The reason all that makes sense is just how concentrated the customers are, right? It's the old rule: the more customers you have, the easier it is to charge them a little more and the harder it is for them to take your margin back. When Intel was winning in the CPU wars, they had, pick a number, 100 million customers, because we're all customers. We all gave them $200 for our Pentium. Nobody cared.

In this case, NVIDIA has 5 or 6 customers that are spending the vast bulk of the revenue with NVIDIA. They're making 75% gross margins, so every one of those big customers should be saying, “It's damn hard to build a semiconductor.” I disagree a little bit with one point Jason made. I don't think that it's easy to build a TPU that can also be rolled out to everyone with all the CUDA support that NVIDIA has.

But even if it's just used internally, and I can save that $20 billion of profit, hell, I gotta look at that if I'm Google. I gotta look at that if I'm Amazon. I definitely gotta look at that if I'm Tesla. So, yeah, I think that's an interesting medium-term pressure point on the NVIDIA profitability story. It's not overnight, and it's not gonna make sense for most people. You have to be pretty damn smart technically to ship a comparable chip. But in a world where you only have 6 customers that matter, having 1 of them say, “I got a better product myself,” is a significant event.

Chamath Palihapitiya

Especially if I'm OpenAI, which is burning even more money than Anthropic, right? I've got to be thinking, “If I could cut 2/3 of the cost of my compute, think about my business.” I go from 1 of the biggest cash-hemorrhaging businesses of all time to a profitable business. Maybe 2/3 isn't quite the number, but I'd be relentless about it.

The bar keeps going up. We talk about it. But ultimately, if I can cut $20 billion, $30 billion, all of it, it's a big deal.

Replay what we've just said in the last 5 minutes. It is the most obvious threat to NVIDIA's business. What is NVIDIA thinking about this internally? Jensen must see this very clearly. How does he respond, and how do they protect their business in the wake of being so concentrated and losing those customers?

Jason Calacanis

Yeah, first of all, there's a reason why they sponsor CoreWeave and the next generation of neoclouds, because they're like, “Those guys ain't gonna build their own chip.” A simple, humble $20 billion market-cap company does not have the capacity to build its own chip. So it suits them to have the market for cloud compute and AI be a little more diversified than concentrated. Anything that can make that happen is in their favor.

To some extent, that's all they can do. They don't have a ton of leverage over Google, and obviously now Google is starting to talk about selling those TPUs to others. So it's a threat. What they would say, and they would be correct, is that for most users, because they have such dominance and such validation of the CUDA software layer, for most customers it's gonna be too much brain debt to switch from the GPU you know and love to something new, right? There's probably a significant activation energy, and that's gonna be true for most customers.

The long tail of NVIDIA customers aren't gonna do it. But you're right, Harry. The threat is that all you have to do is peel off 1 or 2 of those big-margin cows, and you're done. The only thing that perhaps protects them, and this will be interesting to watch, is that the biggest potential alternative customers for the Google TPU are Google's sworn enemies.

Microsoft, Amazon, maybe OpenAI, to the extent they want to do data centers, would be the obvious next places to go. They're the other people who are doing so much compute that it will be worth their while to try and digest TPUs. That's interesting from Google's game-theory perspective. Do they do that, take the capital, or do they continue to just keep it in-house and have a structural cost advantage? I don't have the answer there, but it's outside NVIDIA's control.

I think, Harry, I rambled a little, but to your point, it is, to some extent, there's only so much you can do if you only have 5 customers and 1 of them wants to diversify away from you.

That much…

3. The Customer Concentration Risk

David Friedberg

It's the core risk of Mercor, of Surge, of Turing, of all of these data providers: that they all have 2 customers that are more than 50% of their revenue. The bet that you're taking, or the risk that you're willing to underwrite there, is that it's not a core function of Meta or Amazon or Microsoft, and they won't go after the data-acquisition market themselves because they can just spend the money with these companies and it's good enough.

Jason Calacanis

I actually think we're just ignoring the risk because NVIDIA's numbers are just too good.

Psyched.

Jason Calacanis

No, that's what we're all doing. We're ignoring it. This is systemic risk. Back in the day, Twilio losing Uber as a customer meant 12% of its revenue was gone. This would be much bigger, but when the numbers are there, we're just ignoring it. We're just ignoring it.

You don't wish you invested in NVIDIA 5 years ago? Of course you do. You gotta play the game on the field. Even public-market investors have to play the game on the field, right?

So do you think NVIDIA is overvalued today?

Jason Calacanis

No. The amount of compute—I mean, what did Google's head of infrastructure say this week? Google needs 1,000 times more compute in 5 years than it has today.

1,000x in 4 to 5 years.

Jason Calacanis

Yeah. It's difficult to believe NVIDIA won't be a leader in that time. Oversimplifying where all the chips will go and where Blackwell will go, that's a lot of growth to invest in. 1,000x on compute, if not chips. 1,000x. That's better than most SaaS companies right now. I don't know many SaaS companies predicting 1,000x growth at scale.

David Friedberg

I actually wanna go back to the first question you asked, customer concentration, because I often think it's interesting for people who listen to get a sense of not just the public markets, but how we all think of it as investors. You're absolutely right, Harry. The big question on all the AI compute co-attach bets—and data labeling is a good example—was, there's only 4 or 5 customers here.

The logic you had to use was that in the hypergrowth period, your customers, be it OpenAI or someone else, aren't gonna have time to optimize for efficiency. They're gonna be running fast. In that period of time, you can create huge value, and we saw Scale AI create huge value. Then you're right: the fear would be that when things slow down and people start moving from optimizing effectiveness to optimizing efficiency, those businesses get tough.

As an investor, you're always tempted to do them. I will admit, I was scared of the data-labeling companies, and I was wrong. There was a period of a couple of years where they clearly worked and worked really well. The interesting thing will be whether they continue to work for the next 3 to 4 years if, in fact, dollars get a little more scarce. How do you think about investing in a company that logically only has 4 or 5 big customers?

Specifically on the data-labeling market, this is one where I've interviewed the founders of Turing, Scale, Mercor, and Invisible. I've pretty much interviewed all of them—Surge included. The 1 thing that made me feel incredibly comfortable investing in the category was understanding the specialized data requirements that the large providers need.

I hadn't quite thought about the very verticalized data requirements—whether it's surgical data, whether it's bookkeeping and accounting data—that all of these different players are going so deep into. It's so specific and, in some cases, strange and weird that the large customers are never gonna churn or pull away from them because they are so verticalized.

When I got comfortable with that, I was like, “Oh, I'm okay to take this risk and underwrite it because I don't think they're gonna churn.”

David Friedberg

That's super helpful and insightful, Harry. You're right: if you only have a small number of customers, then at some point they're gonna optimize. So then you get into: you have to have something they can't optimize around. And then, if you have that, the next skill you need to have as a CEO is an ability to play extremely good poker.

'Cause you know, you gotta look them in the eye and say, “I know that your $20 million contract is my biggest contract, but I also know that my data is your most important data.” You're right: there's drama, there's tension, but you wouldn't want to be doing a commodity data-labeling play when they start to get to efficiency.

You look at ASML, which sells primarily to TSMC, and it's the same dynamic. One seller, one buyer, to a rounding error. It's manageable. It's not like you can't build businesses in that space, but you have to have something really unique to avoid them pounding you all over.

The deciding factor—the swing vote in favor of taking the risk in those kinds of deals—is the speed at which the underlying market is moving. That's the get-out-of-jail-free card that says, “Yeah, in theory, you could say you've only got 3 customers; they will grind you down, but if they've just got other shit to do for the next 5 years growing the business, they never get around to it,” right?

It's interesting. So I think that's, to some extent, what's happened in the last 4 or 5 years, both in data labeling and, zooming out, in the NVIDIA business at a higher level. Yes, maybe in some logical world, maybe all of Microsoft, Amazon, and Google should already have had their own TPU equivalent because they should have known to do this, but there were bigger fish to fry for a long time.

In a run-fast world, if you're NVIDIA and you're saying, “But I have the product now; it's ready to ship,” you want to get your compute rolling. You can make money for a long time even in a concentrated market. It's only when it slows down. And that's why, if I look back on the data-labeling thing, Harry, the earlier you are, with more hypergrowth ahead of you, the easier it is to be undifferentiated. By the time the growth slows down, you better either be differentiated or, as Jason was, you better be exited.

Jason Calacanis

All that matters is growth today, right? I mean, even Palantir is extremely concentrated at its scale compared to what we're used to. It's part of the AI world. Palantir has triple-digit customers, doesn't it?

David Friedberg

I want to go back now to the NVIDIA-overvalued question. I phrase it differently. I mean, it gets to the same thing. If you look at today's revenue, it's not overvalued. The P/E on NVIDIA, to give that sound bite that I love, is lower than the P/E on Costco.

We talked last week about the earnings before they came out, and we recognized it was an odd time because the show had come out after. But it all played out exactly as we thought, which was that there was no surprise in the earnings because all the hyperscalers had announced a quarter, a week, or a month beforehand, and they'd all said, “Right now, we are buying more stuff. We are constrained by capacity.” So there was actually no data in the NVIDIA earnings of any significance.

The question of NVIDIA's valuation is effectively a function of the end demand for compute. As long as the demand continues roughly where it is today, NVIDIA is far less aggressively priced, as I said, than Costco or than Cisco in 1999, on a P/E basis. The question of whether NVIDIA is overvalued really translates to: Is the demand for compute, as it manifests in 2025 and 2026, steady-state demand, or is it a cyclical peak and, 2 or 3 years from now, we're not going to be spending $90 billion? That's the primary question.

You're right, the secondary question is the whole TPU and substitute question, but I would argue the first one is the primary one. As long as the hyperscalers and the model providers and model companies continue to invest massively and aggressively in compute, then, with the exception of the substitution risk, NVIDIA's business is safe.

You're really saying, at the margin, is Microsoft going to increase another 30% next year? Is Anthropic going to invest in compute? Right now, all those answers to those questions are yes.

My takeaway on the recent—you can't call them convulsions when things only move by 5%—is not some kind of great correction. It's more that the market's doing a pretty decent job of saying, “All you companies that are investing, who's doing it well? Google, you can invest more. Your stock goes up. Who's doing it badly? Oracle, your stock goes down.”

Rory O'Driscoll

The market is doing what markets are meant to do, which is send a signal. Right now, the signal it's not sending is, “Oh my God, don't invest in anything,” which obviously would be catastrophic for NVIDIA and for all of us because it's 7% of everyone's S&P. The market's sending a signal that says, “Those of you who have good businesses and are investing are doing just fine. Go team. Go Google. Go Microsoft. And those of you who are a bit out there on the risk continuum probably should be thinking about that.”

As a Duolingo shareholder, I'm feeling the pain.

Rory O'Driscoll

Yeah. It's been brutal—70%.

4. Sam Altman Enters War Mode

Before we move on from this core debate on the core incumbents, Sam Altman did a memo internally within OpenAI, and he said that we were in war mode against Google regarding its increasing capabilities and the increasing competition in the space. Does this incitement of war mode ever work, and how do we feel about Sam catalyzing the troops into this proactive state?

Jason Lemkin

I've never seen it work. All the leading public companies say this: “We've got to work harder. We're in war mode. It's AI.” And listen, I've just always wondered: Who are they talking to? Who are they talking to? There must be a handful of folks that it works on. Who really cares enough to go into war mode? They're already working as hard as they want to be.

I've just wondered: I don't think the message is to Wall Street, right? Is it to let VPs know they're going to get fired? It's just something that CEOs want to say, right? They want to tell folks, “It's time. We've got to step up.” I've just almost never seen anybody react to that.

Rory O'Driscoll

War mode as a metaphor for rallying the troops—I don't know if it works or not. I always find it a little odd. Have you been in peace mode until now? Did you just suddenly discover there was a war? It raises all sorts of weird questions.

I also, for what it's worth, don't love the metaphor. Let's be honest, we're a bunch of pampered, West Coast, elite-y computer people. If you want a war, the U.S. Marines are still taking applicants, and you can have a real war. Otherwise, save the metaphor for someone who cares, right? So I'm not a fan of the war mode thing.

But I think, Jason, you're exactly right: What are you trying to achieve? Whether it works or not, in my view, is independent of the metaphor. So maybe the better question, to Jason's point, is: Forget the metaphor, forget the war, forget the “work harder” because you're all working hard. What exactly are you going to do today that you weren't doing yesterday?

That would be an interesting question.

Jason Lemkin

Well, I will say, across every professional experience I've had, from startup to scale-up to VP at a Fortune 500 tech company, nothing happens when you're not in hyper-aggressive mode. Nothing happens.

You think it does, and you get releases out, but nothing happens when you're not in hyper-aggressive mode. You can call it war mode, but nothing happens when you're not in hyper-aggressive mode.

Yep.

Jason Lemkin

It seems like it happens, but just keeping up with the release, with the TPU schedule, with the bug fixes, with the patches—I mean, if you've ever worked in a tech company, especially a software company, fixing the bugs could take all year.

“We can't do it, Rory. I know you want to launch that feature, but we have 12 years of technical debt. What about the OAuth? It doesn't work.” You get endless excuses from the team, okay? Because it's true. Because you have decades of technical debt and 100 features, and you just promised a big customer you'd have this feature that still hasn't come out. And now numbnuts Lemkin or Stebbings wants us to do even more.

So you can't get anything done unless you're in hyper-aggressive mode, like mobile for Facebook. I just don't know if telling the troops works. I've never seen it work.

No.

Jason Lemkin

But you do have to go into hyper-aggressive mode because I think what he's saying is, if we evolve at our current pace, we're going to fall behind. It's just that simple. That's what he's saying: We're going to fall behind.

When I meet with startups, especially when you invest later because your money's already in, right? When they're not in hyper-aggressive mode, I don't want to say I lose confidence in case anyone's watching, but I lose confidence. You're just not going to get anywhere.

I need to see, especially today, when folks are falling behind in AI, right? When you have companies at scale falling behind, I don't care about your talk. I don't want to hear about your pilot. I want to smell that your team is in hyper-aggressive mode. If it is, I think you can come back.

If I don't smell it, you have no chance, right?

What are the signs or smells that suggest hyper-aggressive to you? Founders will be listening to this.

Jason Lemkin

You feel it. The management team has a tension to move faster.

Yeah.

Jason Lemkin

Every single person on the management team says, “We are shipping product faster. I am selling harder. I'm getting on the road more. I am generating more leads.” Every single person on your management team is sweating it, and they're executing faster. You can just see.

They come to your board meeting, they come to whatever, and there is velocity in every area. How often do you truly see velocity increase in a board meeting? Not that common, right? That’s hyper-aggressive mode. Everyone is online on this.

Somehow, this makes a great CEO. This is very hard for a CEO to do. Everything is in an anti-inertia state. Everything slows down, even in an early-stage startup, right? And then when you come in and everyone’s sweating and they’re like, “You know, we shipped twice the story points, twice the software as last quarter, and we’re still falling behind and I’m pissed,” these days, in the area of “Great job,” I just don’t see it enough. That’s what you need. You need to make your team hyper-aggressive. It is an unnatural state.

5. The Consumer AI Battle

If I push you both, who will win the consumer? Is it OpenAI and the app rollout that they are clearly going to do over the next few years, or will Google retain the consumer, the search layer, and the app layer that they have today?

Jason Lemkin

I think Sergey brought back hyper-aggressiveness to Google. He’s clear in the interviews. He just totally said, “I came in, we weren’t even allowed to use our own tools. We weren’t allowed to use our own coding tools. We weren’t allowed to use our own chips.” He’s like, “I got rid of that in a week. I heard arguments, I got rid of it in a week.” That’s what you need.

Sergey had to come back and make it hyper-aggressive. As great as the leadership is, I think Sergey had to come back and make it hyper-aggressive. So I don’t think it was the only point, but you need that. You need that, and most of the startups that I see that aren’t founder-run are unable to get back to hyper-aggressive mode. It is impossible for them.

I’m not saying it hasn’t happened with Satya, but look at our portfolio. How many folks in your portfolio without a founder can switch into hyper-aggressive mode?

So you’re saying Google?

Jason Lemkin

Over— I have to— this is a 20VC, I have to pick between the two.

Yeah. Between the two, will Google retain the consumer with the app suite that they have, or will OpenAI continue to eat more and more and more of Google’s core business, and we will see a much more shared landscape?

Jason Lemkin

Well, listen, I think the same thing most people on X do. I think the Google products are great. I use them every day. I use AI. In fact, I don’t even care whether I use AI Overviews, ChatGPT, or Anthropic, but I actually prefer AI Overviews because I don’t have to leave Google. That’s a great product. So is Gemini 3 Pro. These are great products. You’re not giving me any reason to leave. They’re truly great software.

Mm.

Rory O'Driscoll

I’ll answer this question, and I actually do want to come back to the leadership thing because I want to chat a little bit about that and maybe learn something. But going back to the first question, the Google question, if you listen to the AI Overviews stuff from a while ago, Google six or nine months ago was like, “Oh my God, Google’s dead.” It was all AI Overviews. Obviously, Google executed really well in cloud and with the model, and fundamentally, the core search business has not declined nearly as much.

There was an interview with, I think, the head of search in The Wall Street Journal, and she talked about how search volume is going up as people ask more questions. So it’s clear that the decline of search isn’t going to be precipitous. That’s my clear takeaway from that.

Jason Lemkin

They got that one wrong.

Rory O'Driscoll

Exactly. They got that one wrong, and they’re going to be able to get some search. At the same time, I do believe there has emerged a new and separate consumer category of paid subscriptions for advanced AI, and I think ChatGPT is well-positioned in that place. They have, what, 800 million users, 5% of them are paying. I don’t see all those users, quote-unquote, going back to Google.

I think they’ve carved their niche—a big niche—in the attention economy. I don’t see it as an either-or. Even if the search business slows down in growth, I think ChatGPT can charge and can build a defensible, large, enormously large consumer business—to a minor extent relative to Google and to a more macro extent, to the tune of a few dollars from everyone, right?

I do think the advertising dollars—we’re getting close to the point where the digital advertising spend across Google, Facebook, and Amazon, even before ChatGPT existed, was getting to be a very high percentage of the total ad spend. So if ChatGPT is going to get its share of that, either we’re going to have to grow the pie overall or someone’s going to have to lose a little bit.

Chamath Palihapitiya

Net-net, I don’t think it’s an either-or. I think Google’s proven that reports of its death are greatly exaggerated. I saw in the memo leaked by Altman that he made a reference to growth falling to 5% in a year or two, and obviously therapists call that catastrophizing when you postulate something that’s so awful. This company is growing north of 500% year on year. If it went to 5% year on year, that would be the single most catastrophic growth decline in history.

But in terms of likely outcomes, independent of the human dynamic, the likely outcome for me is that ChatGPT continues to build this compelling consumer product and can command a differentiated market, while at the same time Google Search does not, quote-unquote, “go to zero.” The pie expands. The great American tech Magnificent Seven suck more of the world’s profit dollars out of the rest of humanity. Go team.

Okay. Well, speaking of expanding pies, do you want to add anything, or am I—

Jason Calacanis

No, no, let’s hit pie.

So, speaking of expanding pies—

Chamath Palihapitiya

I did want to go back. The leadership thing is, as I say, for me, because I’m not a good rah-rah-type leader, I was actually interested in Jason’s comment, and I’m reflecting on it because I tend to be more the step-back person, which I think is why I’d be a mediocre CEO, and I’m okay as an investor.

It was a great question. What are the signs? As an investor, when do I look at my companies and go, “God, this guy’s running hot,” right? I’ve been thinking about it. It’s a super question, and it’s generally when, at least once or twice a year, I think to myself, “Mr. CEO”—and I have 1 or 2 in mind—“are you pushing your team too hard?”

Even though I don’t love the war metaphor mode, you are correct. It’s only the wildly intense ones, the wildly driven ones, that really have excellence. I’m reflecting on my very best, excellent companies, and every once in a while, the CEO would just lose his shit. He or she would be so exasperated by the lack of progress and the drive they felt to win that you’d have that tension.

The bigger the team and the more people you have, the more simply you have to communicate. Therefore, you probably use simpler metaphors, and therefore, you probably have to go for some kind of “It’s a war, we’ve got to win,” because that’s what it takes to motivate 5,000 people and point them in one direction.

Jason Calacanis

Well, I don’t know if it’s the 5,000. Thinking on it, I actually don’t think— You said you sometimes advise your CEOs to go easier on your team. I don’t think you have to go easier on anyone great, ever.

Everyone great—every great leader, every great VP, every executive, and they’re rare, they’re rare—is going just as hard or, in some ways, harder than the CEO. The CEO’s job is harder. A lot of executives don’t get how hard it is to be at the top, but in their own domain—CTO, CRO—they are going harder than the CEO.

If this was 2021, people would say I’m being toxic saying this. It’s not, okay? We’re going to 2026. I do not think you can push your team too hard. I think you should push them as hard as the business needs to go, and if they leave, it’s great. It’s terrific because they’re not going to get you there, and maybe that’s what Sam’s doing.

He’s saying to his team, “Complacency has set in. Everyone’s getting to sell 20 million at a time. Turnover is high, and it worked for a while, and now the competition is up.” Maybe he’s talking to 20 people and saying, “I need you guys to go even harder,” and maybe half will leave and half will do it. I think that may be the whole message.

But I honestly think, to founders, my number-one bit of advice is: do not go easy on your team. Love them, back them up, don’t pick at them. Back them where they’re good, let them run where they’re good, but push them even harder. The best ones will always step up. They might cry or lose it for a couple of days, but they will step up.

Chamath Palihapitiya

Or they’ll go, and you’re right, Jason. Generally, in your best companies, you’re exactly right. Some people crash out because they just can’t take it. My point is, how do you know the motor is running at full speed? The answer is, you hear an occasional gear grind, and your job as a board member is to check in on the gear grind and go, “Oh, okay, I can live with that gear grind. It’s okay. They’re doing the right thing.”

Jason Calacanis

Let them all go for 2026. Let them go because they’re not going to get you anywhere. They’re not going to get you anywhere, the folks who think it’s too hard, like the job’s too hard, right?

Literally, in my fastest-growing portfolio company now, at the last board meeting, the CRO stood up and said, “I want to do more than the stretch plan next year.” Calmly. “I want to do more than that,” and gave a data-driven reason why. No one asked him to do it.

You want that out of your team in the age of AI, or you’re just going to fail.

Chamath Palihapitiya

I wonder. I’m going to disagree. You can’t push to the point of being delusional. You can push to just one level below that, right? I actually think that’s the technique: really drive the team to the point where they think, “Hmm, this is not realistic,” and then take it down one notch.

Look, it doesn’t work if you lose your entire team. It could segue into this: you made this discussion about the war thing. As you point out, in the last 2 years since the OpenAI board change, you have lost an entire team at OpenAI.

David Sacks

Well, not all of them. You kept Greg.

Chamath Palihapitiya

Yeah, you don’t want to lose your entire team. So the question is: how hard can you push while not blowing the whole thing up?

David Sacks

You’d rather not create an Anthropic when your team leaves, too, let alone Thinking Machines and all that. You don’t want to force 2 of your best people out and have them build your top competitor.

6. Sierra Tests A $10 Billion Valuation

We said about war mode that there is no category more competitive today, it would seem, than the customer service and customer support market. Brad Taylor, formerly co-CEO of Salesforce, started Sierra and reached $100 million in ARR within 2 years, as it announced over the last week. The last round was at a $10 billion valuation. It raised $350 million from Green Oaks in September. It’s at 100x ARR as of today.

How do we analyze this very fast scaling to $100 million in ARR? Does it justify the price paid, and does it show that customer support is going to be so much bigger a market than any of us previously thought?

David Sacks

I’ll tell you what I can tell you from the street, having a lot of investment in this space. It’s not a criticism of Sierra, Decacon, or others, but the whole category, on the enterprise side, is massively overselling what it can deliver today.

Now, so did Replit and Lovable at the start of this year. Everyone—Replit, Lovable, Microsoft—said, “Give us one line, and we will build you Salesforce.” That is much closer to the truth today. Early in the year, it bordered on fraud. Now it’s becoming reality.

So many people I know have bought next-generation AI support tools and have not deployed them at all, have barely deployed them, or have no AI working. It’s not that it can’t happen. It will. I believe in it, and there are few categories AI can disrupt more than support.

But I will tell you, if you dig deeper, you’ll find that a lot of deployments haven’t even happened, are only half there, are untrained, or are broken. So it’s a lot of getting CEOs and others excited, and this is as oversold today as vibe coding was earlier this year. I’m not saying it won’t catch up. I’m not being cynical. It’s oversold.

Chamath Palihapitiya

It is clear that support, alongside coding, is one of the 2 largest and most obvious markets for LLMs. This is a big, huge category that can totally work.

You start with that, but Taylor is obviously wildly talented, and Sierra has a great name, especially at the high end. I actually think the category is working. I literally did a reference call with someone, and they said, “Look, we evaluated Sierra. We evaluated all these other names.”

Before LLMs, I had an investment in a company that was in pre-LLM customer support. Our resolution rate was around 23%. In other words, we could solve 23% of calls, maybe up to 30%. With LLMs, with these new, modern things, you can solve 60%, which means that on a number-of-calls basis, you can significantly reduce your customer support.

So I think there is some meaningful value there. I’m sure at times it’s overstated, but if this isn’t a market for LLMs in the enterprise, then nothing is. So let’s leave that to one side.

The real question, I think, is: how does the math work from here? It’s $100 million, and it’s trading at $10 billion. What would it take? I just laid it out in my head. You went from $10 million to $100 million in the last year, so you 10x’d. Say you 5x next year, that’s $500 million. 3x the year after, $1.5 billion. 2x the year after, $3 billion.

David Sacks

Wait, when do we have to get to $3 billion in ARR again? Slow down. What year?

Chamath Palihapitiya

And then you go to $4.5 billion and then $5 billion.

David Sacks

Okay.

Chamath Palihapitiya

Over the next 5 years, you’ve got a 10x, 5x, 3x, 2x, 50% growth, and then you slow down a little to 20% growth, and you’re at $5 billion in 5 years.

For context, Service Cloud today, which is the largest cloud within Salesforce, is $8 billion. You’re getting $5 billion out of that marketplace. At that point, if you’re valued at the Salesforce multiple of 5 or 6 times, you’re worth $25 billion, so it’s a 2.5x.

What it highlights is the amount of growth you have to underwrite to make the math work, right? Then you zoom out a level to something we said last week that I believe: when you’re dealing with hyper-growth markets, it’s all about the TAM.

The crude math I just did is that if this company grows faster than any other enterprise software company, in 5 years it’s doing $5 billion, which passes the sanity test of Service Cloud doing $8 billion today. What that says to me is that if all you are is a next-generation software play, you’re probably not going to get that growth, because that would have to come dollar for dollar from Service Cloud.

Rory O'Driscoll

It goes back to what Jason says. The only way this math works is if you eat a huge slug of the labor. Because you eat the labor, instead of the $20 billion software market for services, you eat the $200 billion-a-year services market for customer support agents.

Jason Lemkin

It’s got to go from $100 million in ARR to $1 billion to justify the valuation on its face, right? It’s got to go from $100 million to $1 billion in a year. Can you even deploy in the enterprise at that pace, right, versus self-serve?

We’ll see if they go from $100 million now to $1 billion at the end of next year. We can make a calc sheet.

Has any company gone from $100 million to $1 billion?

Rory O'Driscoll

Well, I think yes. Anthropic, easy.

Cursor and Anthropic.

Jason Lemkin

But how much of that revenue is self-serve, too? There’s a certain amount of human capital you also need to deliver this amount of revenue in the enterprise, right?

Yeah.

Rory O'Driscoll

Agreed.

Jason Lemkin

Because let me tell you, if Brett Taylor walks into a Fortune 500 company and says, “Give me $10 million, and I can replace half your support team,” trust me, he will leave the building with a $10 million contract.

Marc Benioff was great at this, too. Back in the day, when enterprise SaaS was booming, he would go to a big company and say, “What’s your number-one problem, Harry?” Harry would say, “Well, I can’t get this website and this e-commerce…” Marc would say, “Just give me $10 million, and I will get this for you.”

That was the price, and he was magical at it. Brett’s going to be even better, even though Marc’s the best there ever was, because Bret was the CTO of Facebook and the co-CEO of Salesforce. He’s got the best package. No one has a better package than Bret.

He can get $10 million checks, but can he get $100 million next year and deliver the training and FDEs? It’ll be interesting to watch, right? It’s a lot to deliver.

Rory O'Driscoll

I think, Jason, you’re exactly right. Going from $100 million to $1 billion on an API business is very different from going from $100 million to $1 billion in a business where each $10 million contract involves a huge amount of change management in a large corporation in America.

I think the physics of diffusing this technology into the enterprise will be the rate limiter on how fast this company can grow. It won’t be raw demand. It won’t be the talent of the CEO. It won’t be the product. It will literally be whether you can really roll out 100 $10 million customers or 1,000 $1 million customers in a year, where each one of them has its own special sauce, its own dynamics, and its own integrations.

I think there’s a physics to how fast you can grow in enterprise software, even with huge demand, when this amount of change management has to happen. I think that would be the rate limiter here.

Jason Lemkin

Yeah, people who haven’t sold to the enterprise might not realize how true that is. Literally, I think if Bret Taylor could sell $10 million to anybody—

Agree.

Jason Lemkin

It’s not just that he’s so smart or charismatic. It’s that enterprises are literally complex to change. They’re looking for the best person in the world who can solve their problems.

What’s my big problem? If you can dramatically increase the KPIs and support for me, and Bret is the guy, I will give you the $10 million. For you and me, it would be hard to get $10 million. You and I might have to start with a $10,000 contract or $100,000, but literally, I’m not being facetious: Bret can get multiple $10 million checks.

So as goofy as this sounds, the $100 million almost isn’t impressive to me. It’s not impressive because I think he can will it out of the ether with his background. A billion next year with 1,000 FDEs—somehow, he’s got to go hire and deploy that. That, to me, would be magic. It will be magic because they will give him $100 million.

He might even be able to walk into one of the biggest Fortune 50 companies and get a $100 million contract from one, like Palantir. If he promised the moon, they would give him $100 million.

Rory O'Driscoll

The zoom-out you’re right about, and it’s something you learn as you see a lot of enterprise software over time. Big companies and big leaders have big problems, and there are only a small number of people they can take their problems to.

If you look at large technology companies over the last 40 or 50 years, the ones that become dominant in an enterprise wave are the ones who have, one way or the other, projected to the CEOs and CIOs of the largest companies in America: “If you have a problem in this sector, we will make it go away.”

Cisco started off with routers, but over the next 10 or 15 years, it bought pretty much anything you needed for the network.

And I think when Chambers was running this thing, the big-picture value proposition was, “Hey, Mr. CIO, if you have any networking problems, we’ll just buy whatever we need and make it go away. Therefore, you can safely give me $50 million or $100 million.” IBM back in the day was the same thing.

And you’re right, Jason. I think someone as talented as Bret Taylor in the new world of AI can do what I think C3.ai tried to do and failed, and what Palantir is doing, which is walk into a CEO and say, “Dude, you told your board that a top-2 initiative for 2026 is to make XYZ Corporation AI-enabled, and I can help you with that. Give me a $20 million check.” I absolutely agree. I think that is a thing, and it’s frustrating because all the little companies run by ordinary folks can’t do that.

But that’s the magic of being a successful, proven enterprise leader for 20 years. So I agree. I think you can access that kind of business, and it’s a very powerful positioning. I think it’s all about the physics.

To me, the assumption you have to make as an investor with this goes back to a very wise statement that you said, Rory, many shows ago, which I think about literally every single day: the bet within AI is, will we see the transition from human labor to software spend? And if we see that, and we see call centers go and all of the call-center costs go, then it will go to these players, and then we have a $100 billion Sierra.

Rory O'Driscoll

The question is, as is always the case when something is obviously true, valuation expands to recreate risk. In other words, there’s a point at which you do any of these deals all day, every day. The real question is, at 100 times one-way revenues, can you make the math work? And that, I think, is a lot harder.

Are you ready for an uncomfortable question for Rory? He can throw it back at me when we move to Lovable.

Jason Lemkin

Okay.

You mentioned Intercom there, doing, say, $300 million or $400 million in revenue at $2.5 billion to $3 billion in price versus Sierra doing $100 million at a $10 billion price, but the growth rates are wildly different. How do you think about which one you’d rather be in?

Rory O'Driscoll

I think both can make a ton of money, and I’m very happy with my bet in that company. I think they’ve done an amazing job in a way that very few SaaS companies have of transitioning to an AI-first world. And the growth rate of their AI product is, frankly, comparable to Sierra’s, without revealing details that the company should choose to reveal.

So I’m not sitting here— in fact, I feel, broadly speaking, pretty smart having gotten some very nice AI growth alongside an existing SaaS business at a very attractive price. So I feel pretty damn good here. Thank you. But there will be more than one winner as this market segments.

Jason Lemkin

I will say, at a higher level, nothing against this bet, but today, having to support an installed base versus getting to invest just in AI-native customers, it’s a drag. It is a drag. And I was literally with a portfolio company that’s got $45 million in AI revenue growing about 100%. Now listen, that’s not Lovable, but it’s pretty good, okay?

And then it’s got $50 million of pre-AI revenue growing zero. The problem is they are inexorably linked, but they’re not the same product. It kind of sucks. We may change our minds in a year and say, “My God, that install base was the greatest thing ever to leverage for AI.” But right now, it feels like a drag to have to have a couple thousand pre-AI customers to make happy. It feels like a drag.

Rory O'Driscoll

Re-recording my profound opposition to that. I disagree. I actually think every incumbent in these markets has one huge advantage and one huge disadvantage, and I’m going to say it clearly here. I think the advantage is the customer base, the data structures, and access to the data on your existing business, because I think for something like sales or customer support, the truth is it’s going to be a combo package for a long time to come: some automation and then some human agents. Being able to move seamlessly between them both has huge advantages.

I think the disadvantage that every large existing incumbent has is they can’t get out of their own freaking way. Therefore, they can’t leverage the asset they have while at the same time embracing the AI technology. And I think that’s, frankly, something that Intercom did really well.

Jason Lemkin

Well, I think it’s more subtle than that. I hear your point. In theory, it’s an asset, right? In theory, it’s great to have an install base with all of its data rather than a new AI company hoping to have that data in theory. The problem isn’t that. That is an asset, and that’s why I might change my mind in 12 months.

The problem is all the technical debt, all the feature debt, all the features you’ve promised to those 1,000 customers, those 5,000 customers, that don’t give a rat’s ass about your shiny new AI feature. The fact that you have to keep them happy and not let them deteriorate can consume the majority of your engineering and development time. It really can.

I’d love to hear Eoghan’s honest thoughts. I’d like to hear his honest thoughts. I suspect what they did at Intercom—you would know better than me. I suspect they did what you have to do, which is they don’t get so much attention, because there are only so many engineers—100, 500, or 1,050. Your existing customers can consume all of your story points and all of your engineering time, all of it.

You have 10 years of tech debt, and if you’re a new AI company, you don’t have the debt. It’s a huge negative, don’t get me wrong. But the ability to run—when VCs talk endlessly about how AI-native customers are better, I want to gag with a spoon. Sometimes they’re better because they’re 4 really smart people. Sometimes they’re just better because they don’t have 1,000 complaining customers to support.

I’m just saying, right now I feel like it’s a liability, but in a year I may think it’s the greatest thing. Looking at Agentforce, right now it’s early, right? But Salesforce could be the next Google. In 18 to 24 months, we could be like, “Oh my God, Agentforce crushed it. All these startups—we didn’t need them because Agentforce is so good.”

I know it sounds crazy, right? I may change my mind, but right now the startups I’ve invested in with large install bases feel like they have frigging cement shoes.

Rory O'Driscoll

I understand what you’re saying, but again, I would disagree. I think the great advantage—you made a comment about a company that has $45 million of AI revenue and then a whole bunch of customers who don’t care about the AI revenue at all, and that was the revealing sentence.

I don’t think that’s the experience you’re seeing in some of these cases. When you have an obvious—anyone who is running a customer-support organization knows they’re going to be embracing AI, right? And a successful company goes in and says, “We’re going to help you on that journey.”

It’s unimaginable to me to think of a customer-support executive who’s running their business and saying, “No, I’m not going to do any of this AI stuff. I just really like paying people in the Philippines to answer phones.”

Jason Lemkin

But the debt is real, right? And how much you can do for them is real. I mean, I did an interview for G2 with the CEO of Zendesk, and every customer has AI at Zendesk. I’m not an expert in Zendesk, but he was very direct, because they’re private now.

The base version of Zendesk, which you get with AI, is about 20% automated. That’s about as well as you can do without training. And to get to this Fin level of experience or better—or Decagon or Sierra—you’ve got to train the thing for a month or three weeks with FDs and the rest.

So it’s not that they both don’t benefit. It’s just—I think it’s an asset and a hindrance. I think if we look at some of our fastest-growing companies, they don’t have cement shoes. I’m not saying Intercom’s not the exception to the rule, but they just don’t have cement shoes, right?

Rory O'Driscoll

I agree. I think it takes, frankly, really excellent management and a certain—going back to the war-mode comment, which again, I didn’t love, but I’m changing my mind—it takes a lot of product leadership and clarity of vision to make it happen. You’re right, and I think you’re seeing Benioff trying to figure out how to do that, right?

The bigger the organization, the harder it is. I think the team at Intercom has done a really nice job of doing that, right? It’s hard.

Jason Lemkin

Well, let me put it differently. Intercom did it. Let’s stipulate Eoghan did it, and it’s great. This is why I think most of our unicorns will fail, because you’ve got to be as good as or better than Eoghan to make this transition. Most of them are going to fail.

When I look at so many of these B2B unicorns, it’s so hard to do both. It’s so hard. I’ve got $200 million ARR, Rory. I’m growing 20%. I know I have to do AI, but I can barely get the team to do what I’m doing. Unless you have this aggressiveness, war mode, great CEO, there’s just no way you can do both. There’s no way. Your typical PE company can barely get a release out a year.

Rory O'Driscoll

Agreed. I think there are 2 things that stop you. And I think, as is usual when you and I talk, Jason, you emphasize one and I emphasize the other, which is okay.

The 2 challenges you typically have: one I call the management challenge, which is what you emphasize. You’ve got all this old stuff, you’ve got to do this new stuff—it’s hard to do both, so you’ve just got to be a good manager. But then the second thing I would argue, the product thing, is there’s got to be some obvious linkage between the 2, right?

If you’ve got this SaaS company in the old world that does X and there’s no obvious X-plus-AI equivalent, you’re just on an island. Then deciding, “Oh, I’m going to do something new in AI,” you’re screwed.

If all you have is, “I have customers who might want to buy AI in the future, and they buy SaaS today,” you’re screwed. The only time you have even a chance to do it is when both organically go together.

I’ll give you another example of that. I think Gong is a very interesting company that was pre-GPT AI, right? Call recording. I think they’re doing a decent job of navigating that terrain and adding all the LLM stuff on top.

But it’s not because they’re management geniuses. Maybe they are. It’s more fundamentally because there was an obvious path from here to there. In some other areas where the SaaS thing doesn’t have an AI equivalent, you’re on an island, and then you’re right: there’s nothing you can do. You could be God’s gift to management, and you’re screwed. I’m trying to think of examples of that without throwing anyone under the bus, but there are deals where you look and just go, “Hmm, I’m not sure we need you in this AI-first world, baby.” That’s not good.

7. Lovable Tests A $6 Billion Valuation

You said the fastest-growing companies are tied to AI-native customers. One that’s really tied to this: Sierra hit a revenue milestone. Does it justify the price? Lovable hit $200 million in ARR, or 2X what it was in just 4 months. There’s a rumored $6.3 billion round, in terms of new price valuation being $6.3 billion. Is that justified when you see Lovable’s growth rate hitting $200 million in ARR?

Rory O'Driscoll

I don’t know, but I will tell you, for folks who are critics, this is an old SaaStr lesson that I really think we have to follow in the age of AI: you’ve got to segment your customer base.

What I mean is, I assume at the low end the churn approaches 50% or higher, okay? The one Menlo just did in music—what’s it called? We used it.

Suno.

Rory O'Driscoll

Suno, right. They only retain 20% of their customers in a year at that level.

But I just did a presentation at Replit that showed all my apps at an all-hands, and while I was there they closed a pretty large 7-figure deal. The point is, there’s no way that’s going to churn as a multi-year, 7-figure deal in anything less than a couple of years, mathematically, is there?

What I would almost want to do is take Lovable and Replit and segment them and say, “Listen, there’s a high end here that’s probably got 140% to 160% NRR. I’m pretty confident of it. There’s a classic mid-pack that, with upsells and stuff, my guess is approaching 100% retention. Then there’s the part at the bottom that’s worse than we’re used to. It may have 40% retention or 30%, but that’s not unprecedented.”

Almost every mobile subscription app, as we know from RevenueCat, has 20% or 30% retention. I would almost want to not only segment the bar chart, but almost draw a black line through the bottom if I were investing. I’d just call that marketing spend. That bottom is just marketers. They’re just folks getting the word out, and I’m betting on the next 2, right—the 100% and the 160% NRR.

Even if that’s half of Lovable’s, I think it’s a decent bet, right? As these guys grow, their NRR will go up and the churn will go down just because, for classic enterprise reasons, those big customers are going to be stickier. They’re just going to be stickier.

I’m just always thinking about the opportunity cost of cash and where I put my money to make the most money. To Rory’s brilliant—

Rory O'Driscoll

Well, you’re a trader.

To Rory’s brilliant point on Harry, that’s great, but what about me? It’s now worth more than Wix. $2 billion in ARR. Admittedly, the growth rate for Wix is 14% year on year, but they have Base44, which is growing very, very fast. It just announced it hit $50 million in ARR. So you have to believe that either Wix is very undervalued or Lovable is very overvalued.

Rory O'Driscoll

It’s a tough one. You know what the tough part is for all public company leaders? It shows you’re not getting credit for checking the AI box. Hooray to Wix: you went from nothing to $50 million in vibe coding. If that means you’re this percentage of Wix, or of Replit’s level, you should be worth a couple of extra billion. The market’s saying, “No, it’s not enough.”

As a public company, you’re being judged as the public company. You’re not being judged as Sierra. It’s tough because every public company is out there hustling its AI story. I’m not sure it’s working at all. I think it’s necessary, but not sufficient. You need to be Palantir. That’s what the public markets want, and they’re saying Wix ain’t Palantir. $50 million’s not enough. Show me $500 million.

Jason Lemkin

But I would caveat that all you can conclude is that right now the public markets aren’t giving it full credit, which is different from saying the strategy is right or wrong. As we just discussed, everyone was dumping on Google 9 months ago, and now the stock’s up 2.something X, right? Markets are fickle things. They change their mind.

The question is, is it the right strategy for Wix? If you’re in the website-building business, which is what they have been, it seems almost inevitable you have to add this. Otherwise, you’re not relevant. Agreed? That’s kind of step 1.

Then the question is, directionally, they’re doing the right thing. Let’s start with that. You can measure how well they’re doing it in terms of adoption: what percentage of their customers are using this product? Is it off to the side and just a game, or is it a core part of their product? You’re measuring, with industrial logic and not financial logic, how well they’re doing.

Then you can start saying, how do you compare that bet to the Lovable bet, right? That’s the way you have to do it. It’s a $2 billion business versus a $200 million business. It probably has better churn and retention, but only a small percentage of it is AI.

On the other hand, you have Lovable: it’s all AI, but you have massive churn. It could well be that the public markets are undervaluing one and the private markets are overvaluing the other. In the short term, you know the old cliché: in the short term, markets are a voting machine, and in the long term, they’re a weighing machine.

If Wix pulls off its strategy, it’ll get valued for it in the end. Ditto Lovable. If they don’t—if they can’t overcome the inertia, if they just tick the AI box but don’t actually make it core to what they’re doing—then you’re right. Fast-forward 5 years, and it’ll be Lovable is the king of website generation. Who was Wix? They didn’t make it through the turn.

Guest

It’s just a reminder to founders and everybody: $2 billion in ARR, growing 14%. As Harry and I do this, the market cap is $5.24 billion.

For Wix.

Guest

Yeah. Generally, once you descend into low growth—and that’s different for public companies than private companies—I mean, if you’re worth anything, you might be worth nothing as a startup. You’re worth in the 3 to 5X ARR range, and this isn’t even that. This is less than 3.

I just got back before this. I was over at one of our joint LPs at Horsely Bridge, and we were talking about the pulse. I’m like, “Man, it’s the best of times ever in startups, but going public and IPOs are tough out there.” This is 2.5X revenue for Wix. I’m not saying it’s worth more on a DCF basis or whatever—Rory can help me there—but it’s brutal to be worth less than 3X ARR, right?

Semrush just got—I mean, you look—oh my God, it got acquired by Adobe for $1.9 billion.

$1.9 billion.

Guest

Yeah, that was 3 times revenue with a 100% premium. I think a 100% premium, right? It doubled. That’s with a massive premium because Adobe paid up. It’s a bootstrap company, so it’s a fun story, but it’s a reminder to founders who are hiding, or who got the 2023 message that all that matters is getting profitable, that it’s just a brutal world when growth slows.

It’s a brutal world. Even if you inject AI, and even if you add Base44, Rory’s right: it may be a great story in 2027, right? You can draw the math, but it’s not a magical solution today. It’s brutal, and it’s brutal across our portfolios that have a little bit of AI that hasn’t led to Lovable growth. It’s brutal.

Guest 2

What are they doing? Wix is doing about $2 billion a year, and it’s $50 million of Base44. So, yeah, it’s 2% of revenue. At 2% of revenue, it’s a rounding error. Discard it.

The interesting thing is, if they could upsell 10% to 15% of their business to that, where they’re adding $200 million to $300 million, you’re right, Jason: the multiple for companies doing 10% or 15% is pitiful. The multiple for public companies doing north of 25% is compelling.

Yeah, it could double if they do $250 million next year, right? It could more than double their market cap.

Guest 2

Yeah, more than double your market cap. They have to be saying to themselves, “Can I drive 15% penetration of this AI product across my customer base? If I can make that happen, it’s great.”

Yeah, let’s be clear to our public traders, Harry. If Base44 really can ape Lovable and Replit, everyone should pile into Wix because, all things being equal, the multiple should radically inflate. If it can do $250 million—and this is ARR, I guess this isn’t GAAP—but if it can do $250 million of ARR next year, that’s massive over growing 14% on $2 billion, right? It’s massive.

You should pile in. You should put as much of the 401into Wix as you can. This is the great undiscovered public company.

It is, unless the public markets continue not to give you credit for it, as they’ve not given you credit for the 0 to $50 million in Base44 that you’ve just done.

Guest

No, we’re always making the point that you go into the intermediate tier, okay? If you look at multiples, you go from the 3–4× tier, and then actually, in the public companies, you get a really good deal if you’re in the middle tier. And if you’re in the top tier, like Rubrik and Palantir, you get the deal that Cliff from Canva wanted. He wanted that deal because it’s better than the private market.

But I think that middle one is still around 8–12× ARR, this middle group in the 20s. So getting above 20 as a public company is easier said than done, but it’s so worth it. You should buy anything you can to get yourself into the 20s.

Guest 2

I agree, because, Harry, I just reject the whole “give you credit” thing. In the short term, maybe the public markets, quote-unquote, “don’t give you credit,” but I dislike that expression because it implies that it’s some merit thing. In the end, everyone’s just a set of discounted future cash flows. If you point the revenue line up and the P&L line goes up with it, in the end, the market will rerate you and give you value, because that’s just the way capitalism works.

But sorry, if you look at something like Palantir, which is detached from rationality and detached from a DCF model, then it proves you wrong.

Guest 2

No, in the short term, things can be wildly wrong, and one of 2 things will happen. It’ll grow at 70% for the next 15 years and grow into its market cap, or the price will correct. In the short term, prices are often wildly wrong in capitalism, but in the end, it evens out. My point is this: you can’t chase the pricing. You have to run your business, and in the end, the markets will catch up with what you’re doing.

Guest

Well, I think, to go to Harry’s point, Rory, and I think you’d agree, I just don’t think the market believes Base44 is going to accelerate Wix, right? In my little tier on SaaStr AI, if you’re growing less than 20%, a cherry-picked basket is at 5.1× ARR. Wix is lower. It’s not in my basket. From 20% to 30%, it’s 11.8× ARR, and then the 30% plus group, which is Rubrik, Palantir, and Figma, averages 23.7× ARR. But Palantir and Rubrik push that up.

You want to go from 5× to 11.8×, but the markets are saying, “We don’t buy it.” Either we don’t buy it, or possibly we don’t care. “Show me the money. I don’t care.” Look at NVIDIA.

Well, are they just saying, “When you do, call me, and we’ll give it to you”?

Guest

Yeah, call me when I see it. Yeah, call me next year.

Guest 2

I totally agree. I think markets pay up for growth, and it’s why all these people who, in 2022, were like, “Just get profitable”—they’re wrong. It’s a necessary but not sufficient condition in the end. We’re in the growth business, and you’ve got to be growing.

So the question gets back to: can they make that happen? It’s funny you mentioned Semrush there, which, for the listeners, is a company that does search engine optimization. It’s a bootstrap company that’s been around a long time, was public, and was pretty much a sleepy company. Adobe just recently bought them for—you know, what was it?—about $2 billion, Jason? Right? Yeah, $1.9 billion.

My guess is—and it’s funny, it’s top of mind—my guess is that was because there’s this whole new emerging market now for LLMs, you know, answer engine optimization—

Well, GEO.

Guest 2

GEO.

Guest

Yeah, that’s the reason they bought it.

Guest 2

Yeah, yeah.

Yeah, that’s the reason they bought it.

Guest 2

Exactly. Answer and basically optimizing the equivalent of your search results, but on LLMs, which I think is a great market and one we’ve looked at. And clearly Adobe said, “Let’s buy the old-school player, maybe allow us to parlay entry into that new market,” which was just super interesting.

I remember thinking that, because I’d been looking at that market and hadn’t ultimately been able to get an investment done. I remember thinking, “Hmm, if I was a public investor, I’d look at Semrush, I’d buy one of those GEO products, and I’d try and do the same trick of jamming it through the channel.” At 2 times runway revenues, it wouldn’t take a lot to make a pop. But lo and behold, Adobe got there first and said, “No, we’ll take that.”

Well, the interesting thing is, what they said publicly was that they didn’t buy it to jam it through the channel. They bought it because it was the number 1 thing their customers were asking them for.

Guest 2

Yes, agreed.

Obviously, I’m dating myself, but when I was a VP at Adobe, remember, the Marketing Cloud is a huge amount of revenue, and their typical customer is not a 20VC portfolio company startup. It’s a pretty mature CMO that’s trusting Adobe to execute their marketing strategy, and they’re worried about LLMs and GenAI. They don’t know what to do.

I know Harry’s invested in it, and I love Harry. I actually think this is a horrible category, but the demand at this moment is just like a lot of things now: “I need an SDR. I need a GEO solution.” So the demand is off the charts, so Adobe needed a solution today, and they bought what they could get, right?

But I think it’s a horrible category too.

Guest

Okay, you agree with me?

8. The GEO Commoditization Fight

I am actually totally with you. My bet is this company, Peak.ai, is a complete founder bet. To be fair, the traction’s insane. I mean, it’s like 15× in 3–4 months.

Guest

Yeah, the demand is off the charts, right?

So demand is off the charts. But what you’re seeing is the commoditization of the pure discovery and analysis segment, because there are really 2 segments. There’s discovery and analysis, where you rank, and then there’s, “Hey, what you should do as a result, how you can improve it—”

Guest

Yeah.

—and then we’ll do that for you. And you’re seeing the commoditization where you’ve seen Spencer at Amplitude release their product, which now is the same.

Guest

Yeah, they vibed it.

And so my question to you, Rory—which is weird, given that I’m an investor in this, but Jason and I agree—is, why do you not think we’re going to see the commoditization of this category, and why do you like it?

Rory O'Driscoll

I’ll tell you why. I think commoditization is a slur, with all due respect—not a useful word. It’s an implied statement of wrongness, of economic badness about it.

As I remind people, oil is a commodity, but as the film series Landmine tells me, we make $3 billion in profit every damn day making oil, right? So I don’t think “commoditization” conveys something.

I like that market because I think what you see in markets that really work is a wedge product that has high urgency to spend, which means in the near term you get explosive growth. Then, over the medium term, there’s a set of expansion opportunities beyond the original point product that allows you to grow with your customers and expand, and I think that’s exactly the case here.

Discovery up front is the core need, and you’ve got a bunch of companies like Profound and Evertune and all those guys doing that. Then you’re going to add on, over time, content generation and everything it takes to make sure you show up, and I think there’s just a lot to be built on top.

Jason Lemkin

Well, I’ll tell you—for what it’s worth—sorry, what’s your portfolio company called?

Peak.ai.

Jason Lemkin

Okay, Peak, I love. It’s an exception to the rule. There is a category of snake-oil AI, of which GEO is one, and I’ll tell you why it’s snake oil, and it’s going to die in ’26 or ’27.

I’ll tell you why it’s snake oil. Just bear with me, okay? SaaStr itself—we’re not as big as 20VC, but we’re multichannel. I get about 5 million visitors to our blog—about 5 million views a year—from SEO. Traffic is up 50% this year, and our SEO is down 8%.

I’ve tried all the tools. I don’t have time. I don’t have a team like Harry’s. My team is all agents now. It’s shrunk, okay? But I care enough about SEO that I’ll try the tools, especially in self-serve. Nothing’s frigging actionable. Nothing’s useful to me.

Now, when I was at Adobe, I would sit in these meetings, and the marketers would come in and be like, “Well, let me tell you how we did this year.” We had 7 trillion impressions on Facebook. It was just performative metrics.

I get why everyone’s going to buy a GEO tool, okay? I get why you can sell $50 million, $100 million of this, because you need to walk into the meeting and show what’s going on. I cannot find a single thing that’s actionable. Putting more things on Reddit doesn’t help me. An AI tool to write content? I have 10,000 pieces of content.

I know maybe I’m an extreme example, but I’ve tried every tool that I can, and I literally cannot find 1 thing that’s actionable. I’ll tell you the really bad sign with this category—with AI. If you have to put in your credit card immediately, this is a bad sign, a terrible sign today. If your AI is good and you don’t need FDEs and a massive army, give me a few credits.

What’s the music one again? Suno?

Suno.

Jason Lemkin

We use it all at SaaStr; I forgot. So I went to use it again today and forgot I used it. They gave me another 50 credits, and I upgraded, but I did 3 songs for free.

How come I can’t GEO for free if it’s so great? How come I can’t GEO for free? This is a terrible sign. If I can’t try your AI app for free, you’re a fraud. You’re trying to get my credit card before you can provide any value.

And if you look at all the ones that are exploding, half of them—not the Mercurers, but half of the ones we talk about—are massive PLG plays, right? You can’t take their credit card in 60 seconds and grow at this rate.

You can’t. You gotta let them do a couple of Lovable prompts. You gotta let me do a song. You gotta let me try ChatGPT for free. And so I think all these scams are gonna make a lot of revenue. It’s like the old SDRs pre-Claude 4. They’re all terrible too, as you know. They didn’t work. And so we’re gonna see a bunch of things that have massive budgets and don’t work, and they’re just gonna churn.

I think GEO’s a scam.

I’m gonna make a bet here, Rory. I’m gonna put a $5,000 bet on that Peak is gonna be the one, okay? Five grand.

Jason Lemkin

I’ll believe you. I’ll even help them if I can, and I want to learn.

Maybe.

Jason Lemkin

But there might be one.

I have introduced them to the biggest CMOs, the biggest, hardest companies. They fucking love Peak. They love Peak.

Jason Lemkin

Yeah. Do they even know what they’re talking about? I find a lot of CMOs don’t even know what they’re talking about in AI.

Rory O'Driscoll

I disagree, Jason. Genuine comment here: I understand your point about how actionable the information is, but look, you were at Adobe Marketing. Stepping back another decade before you were there, I was an investor in Omniture, which is the company that became the Adobe Marketing Cloud when they bought it in 2008 or 2009, right?

Jason Lemkin

Yeah, and there’s some snake oil there, too.

Rory O'Driscoll

But you call it snake oil. Let me encapsulate the point you’re making, because there’s some truth to it, but then let me make the counterargument. What you’re basically saying is, at some level, these tools just tell you how you’re showing up on the LLMs, just like way back in the day, analytics told you what was going on in your website. You’re like, “What do I do with this information?”

The thing is, you have to know the information, especially in a larger corporation, even if initially you can’t do anything with it. Step 1 is, if you’re the VP of marketing at a large corporation and your CEO comes in and says, “I typed in L’Oréal into ChatGPT last night, and they said 10 mean things about us. What the fuck?” You better have an answer to that, right?

Initially, for that 1st year, just giving that high-priced VP of marketing an answer is worth 30 grand, because it’s hard to explain. But the 1 thing I’ve internalized about all these media types is every marketeer knows, in the end, you just have to go where the people are, and you have to show up where the people are.

When the people went to Yahoo from TV, you had to show up there. When they went to Google from Yahoo, they had to go there. When it went to Facebook, you had to go there, right? The people are using answer engines, they’re using LLMs, and you have to know how you show up there. It’s different. It’s not like you can advertise.

Jason Lemkin

Okay.

Jason Lemkin

But they’re still selling snake oil when it’s unactionable. But you’re right. I have no doubt you can do 50 million here very quickly.

Rory O'Driscoll

But you’re right. What’s going to have to happen—and all these companies are gonna do it—is you have to figure out what you do with that information and how you make it actionable.

Because let me tell you what they’re not gonna do, Jason. No marketeer on the planet is gonna say, “I now know I show up shitty on LLMs, but it’s not actionable, Mr. CEO, so we’re just fucked. Settle in and die.” No.

The CEO’s gonna come in and say, “I don’t give a damn how you do it. You figure out how to get us on the front page of ChatGPT saying L’Oréal—or whatever product it is—is awesome, or you’re gonna lose your job.”

Jason Lemkin

I agree with you. Maybe just 1 point. I think there is a billion-dollar ARR opportunity here. I think there’s a billion-dollar opportunity in AI sales. In marketing, I haven’t seen it yet, and it ain’t GEO.

If you do all of this for real, I’ll give you 100 grand, and if you actually increase the number of qualified visitors to my website by 50%, everyone in the world will give you 100 grand. If you can do the Cursor, I know we all hate the “Cursor of whatever” pitch. A lot of folks say they built the Cursor of marketing. I will buy it. I have not seen it yet.

As soon as they do, this will be a billion-dollar ARR opportunity in 36 months, but the snake oil isn’t gonna do it. Little tiny point solutions that get you 5% of the way there—Amplitude clones—aren’t gonna do it. But there is so much money for actually making it actionable. There is so much money. Everyone’s SEO is down.

Rory O'Driscoll

There are gonna be, in my opinion, 2 levels of actionable. There’ll be actionable right now, where it’s like, “Oh my gosh, this is what the LLMs are saying about you.” Figure out where they’re deriving that information, which is typically via the search that the LLM does, and then figure out how to get nice things said there. That’s kind of the actionable today. And you’re right, that’s loosely coupled at best.

But look, the big shoe that’s gonna drop, and it goes back to the 1st conversation, is when ChatGPT starts allowing advertising. Then the dynamics of this business change a lot, right? If there’s gonna be advertising, there’s gonna be monitoring of advertising, and then you’re right, the wall of money is gonna hit here. People are gonna wanna know how to measure it, how to track it, and how to influence it.

So I think that’s the big-picture bet here. Again, I like the space. I take on board your comments on the product, but go.

When ChatGPT allows advertising, Rory, what happens to the players in this market? Are they hurt or are they helped?

Rory O'Driscoll

Look, you can imagine a world where you could go either way. One of the interesting things about what we talked about with Semrush is that the outcomes in search engine optimization haven’t been amazing. Semrush is actually the biggest outcome at 2 billion; all the other stuff, like website analytics and email analytics, had much bigger outcomes.

So you could argue that if the ChatGPT version of advertising is similar to Google, then the platform gets all the money and doesn’t allow a ton of ancillary products to thrive, and maybe it’s only a so-so outcome. Maybe that’s the case. A lot depends on what ChatGPT implements between advertising and commerce and how much of a role is left for ancillary players. That’s what you don’t quite know at this stage, right?

You can argue that the midterm question for these companies is whether there’s some element of platform risk. Will they allow enough of that? On the other hand, it’s a complex enough problem that if you’re a large corporation spending a lot of money to improve how you, quote unquote, “show up in AI,” intuitively there’s gonna be spend there.

Will it be 200,000 per company or 2 million per large corporation? I don’t know. That’s the risk in the deal. Obviously, you can tell from the way I’m muttering, I don’t have clarity on that. But intuitively, I go back to my 1st principles. If all the people are showing up in the LLMs, then all the advertisers are gonna wanna show up in the LLMs. And if you can help them get there, you can probably clip some of that money. That’s the very unsubtle thesis.

9. IPO Liquidity Remains Tight

Final 1 before we do a would-you-rather, Rory, because I know you love that. Where do we wanna go? We’ve got Figma, obviously—the IPO price and where they are now. Oracle down 40%. We can do Kalshi’s new round. Any preference there?

Jason Lemkin

We can talk about Figma. Maybe we should talk about it; I just feel like it’s a Debbie Downer. The Figma IPO was so exciting, right? Everyone was captivated by it, almost beyond what it deserved, right? Almost at a consumer level, almost at a Google or Facebook level.

For it to be a broken IPO, at least for a while, it’s just a bummer. I know we talked about it a little bit last week, but it’s a bummer given the feeling that liquidity was back and IPOs were back. Maybe Bill Gurley thinks it’s great because they got a good deal, but a super-active IPO market is great for everyone from VC down to founder. It’s just a big plus. Having a meh IPO market is not gonna help any of us.

Rory O'Driscoll

I don’t think it’s meh. I would argue that it goes back to what I said earlier: voting versus a weighing machine, right? The professionals all looked at the stock and said we should transact at 35, which is roughly the same valuation Adobe offered 2 years ago, which totally makes sense. You generally bid forward 1 or 2 years in an M&A.

Then all the retail madness took over, and it was valued at 100 and something. Now, over time, the voting stops and the weighing begins, and now it’s pretty much priced at exactly what they thought it was worth 6 months ago. I would argue that, in the end, the markets are efficient.

It’s a great outcome. Lighten up, everybody. It’s a $17 billion, $18 billion market cap for an awesome company. Yeah, you’re right, the process has been a bit of a Debbie Downer, to your point, Jason, but intrinsic valuation wins out in the end. Capitalism works.

Jason Lemkin

It is a reminder, though: if you turn down a deal, you better want it to go public, because it’s not just down from Adobe. You’ve gotta take in dilution and time value of money, right? I tried to write it up on SaaStr. I think it’s like 30% or 40% lower than the Adobe deal, adjusted for time, dilution, and risk—30% to 40% lower. So when you say no, you better...

Rory O'Driscoll

And granted, they didn’t say no; they said yes.

Jason Lemkin

They said yes, and yes, and yes. Yeah.

Rory O'Driscoll

So who knows what really is going through everybody’s minds, but it’s just tough to not be above that number. We’ve already forgotten about Wiz. It seems like 3 generations ago, right? It seems almost quaint compared to Cursor, but maybe it ends up being one of the greatest deals, smartest deals of all time.

Jason Lemkin

To be clear, you’re talking about the Wiz decision to sell for $32 billion, which deal is still pending.

Rory O'Driscoll

Still pending, yeah.

Jason Lemkin

It’s got U.S. regulatory clearance now, unlike poor Figma, who didn’t. Or indeed, much worse, poor little iRobot, maker of Roomba, which didn’t get clearance and then pretty much went bankrupt because they couldn’t sell their company. So yes, antitrust has had an invidious influence.

Rory O'Driscoll

It just would be nice if everyone who needs to IPO could IPO next year, and I’m not being facetious. It would be nice if the market was overheated, if it didn’t quite make sense, if Figma was trading at twice its IPO price, and it could actually lift some folks that were a layer below Figma.

It would be nice if folks at $200 million growing 30%, who are still in that second tier nominally, could calmly IPO next year with high-demand, oversubscribed IPOs and just free up liquidity in our portfolios. I’m not being facetious. It would be great if we just had a little more froth in the market. It would be very helpful.

Do you think we are going to see any opening of the IPO markets next year?

Rory O'Driscoll

I don’t think we earned it this year. I don’t think we earned it. I don’t think we have so many companies better than Figma or Netskope. It’s a high bar.

In IPOs, it tends to be Pavlovian. People do things when the last thing they did felt good, and Jason’s right: these IPOs don’t feel good, so it won’t be easy to have more.

Now, the way you make things easy is price. Price is one way to make things easy. The other is the kind of companies that go public. So, if the overall equity markets stay roughly okay, then there are lots of companies that can choose to go public if they want to because they’re above the bar. The question is, will they want to? That’s more a function of your individual company decisions on whether they want to or not.

But I’m with you, Jason. It would be great. A more wide-open window would be good for many, on many different dimensions.

Rory O'Driscoll

I just feel like in venture we’re acting as if these liquidity windows are wide open, and I get challenged, but I don’t see it. I don’t see the IPO market wide open. I don’t see billion-dollar M&A deals from PE firms and everyone happening every week.

The limited visibility I have on billion-dollar M&A in my portfolio or with friends—it’s happening, but it’s stressful. No one’s sitting there just writing, throwing billion-dollar chips into the middle of the table because it’s the age of AI, like in some ways they did in 2021.

There is more stress in liquidity in the system than we think with these OpenAI and Stripe secondaries. It’s just a bummer we didn’t de-stress it this year, but that’s our job. I mean, we still have our fees to get us through these rough patches, right?

Jason Lemkin

But, Harry, I’m not going to let that happen. No, I mean, look—

We don’t take fees.

Rory O'Driscoll

You don’t take fees. That’s good.

We don’t take fees.

Rory O'Driscoll

That’s good. It’s success only at 20VC, right?

Yeah.

Rory O'Driscoll

It’s a 0/30 fund. We go big on the carry.

Rory’s the same. Rory told me they don’t do fees either, so I just follow him.

Rory O'Driscoll

It seems dated, doesn’t it?

Yeah, no fees.

Jason Lemkin

I mean, look, in the end, for most normal fund sizes, despite the cynicism, the truth is you make your big money on carry, A, for all the obvious reasons, and B, it’s a tax advantage. So yes, it would be really great if the window opened.

The scary thing, going back to linking it to the AI thing, is that with every year that goes by, the probability is that the next crop of IPOs will be AI-first companies from 2022, or pre-2022 companies that have clawed their way into AI land. The probability goes down that those 300 or 400 companies that are unicorns from before 2022 and aren’t making the transition will ever get out.

Every year that goes by, as the tech debt mounts and as the new world order becomes clearer, the probability of most of those companies getting out has to decline, and that’s the scary thing.

Rory O'Driscoll

Do you think you can maintain a 2021 mark if you haven’t seen massive AI progress? How are you thinking about it as we come to the end of the year at scale?

Jason Lemkin

It’s very hard to imagine any 2021 mark that’s been maintained that hasn’t been validated, or come anywhere close to being validated. We wouldn’t have, and we haven’t.

Rory O'Driscoll

Do you think everyone—they’ve all been marked down?

Jason Lemkin

I think they have. I regularly, for my bedtime reading, eyeball the list of unicorns by year, and you just look at it and go, “Oh, that’s good, that’s good. Ooh, no, no, no, no.” It’s like, yeah, not even close.

I think there’s enough data from the various surveys to say that the secondaries might not be representative, but, yeah, if you have a ’21 mark that hasn’t been validated since then, if you’re trading at 6x revenue and you’ve grown into it, then yay, you. If you’re still trading—

Rory O'Driscoll

Yeah.

Jason Lemkin

—at 20x revenue, you might want to think about it.

Rory O'Driscoll

Do you think 6x is okay if you’ve grown into a 5x or 6x multiple?

Jason Lemkin

It’s growth-adjusted. It’s all the things you talk about. It’s growth—look, the truth is, if you’re in a boring—

We’ve seen it in Semrush at 3x or 4x. You cited Wix at 3x or 4x. On the other hand, with 25% growth, you get to 12. The interesting thing is how fine-grained it is. Small percentage points of revenue growth here, between 15% and 25%, have massive consequences in terms of value.

That’s actually a whole theme, as you talk to CEOs: how easy it is to move from not great to awesome with just a little bit of reacceleration.

Rory O'Driscoll

You’re right, but you know why it’s so much harder than it looks? Because if you’re growing 15% or so, or in the teens, probably the majority is from price increases today. It’s fake growth. It is price increases or stuff jammed down the channel, and maybe it’s a couple percent from new logos.

But if you’re increasing effective pricing 8% to 9% a year, even with some retention issues, the truth is it could be half the growth, right? It’s not that that doesn’t count; it just makes it much harder to get into the 20s.

Mm.

Jason Lemkin

No. In the end, exactly. You’re not going to get into the 20s on price increases. You’re not even going to get into the 20s probably on NDR. You’re going to get into the 20s because new people want your product.

Which is why, going right back to Grove and all, if you cut R&D in the downturn, you’re probably screwed. If you don’t have that compelling second product that you built in ’22, ’23, ’24, you’re just in a tough place.

Okay, we’re going to do a quick fire.

Jason Lemkin

Okay.

Rory, this is your favorite. Would you rather? Yeah? Would you rather be in Wix or would you rather be in Lovable?

Rory O'Driscoll

That’s a good question.

This is from a make-money perspective.

Rory O'Driscoll

Oh, just to be clear, I’ll go with Lovable at the margin. Both companies are interesting.

I don’t have clarity on how Wix’s vibe-coding product rolls out across the rest of the organization. If I had clarity on that, I would take Wix all day, every day, because I think you can make a 4x or 5x that would be liquid. But I don’t have a thesis on it.

You cited my Intercom investment. I had a clear thesis there. I don’t have it in Wix. If I spent the time, maybe I would, but in the absence of data, it’s presumed innocent. In the absence of data to the contrary, the prior is that the AI-first company has the edge in terms of growth.

So I’ll go with Lovable at the margin, but I’m a bit nervous at $6 billion, just to be clear. A bit.

Jason Lemkin

Well, listen, I watched your interview with the Base44 guy. He’s pretty good. We knew he was good, right?

Rory O'Driscoll

Did you think it was good?

Jason Lemkin

Yeah, he’s pretty good. I feel like I know this space pretty well now.

Rory O'Driscoll

Mm.

Jason Lemkin

Listen, of course, he’s a founder. He should know all this stuff, but his fluidity in the space and his knowledge of where it’s going to go and how to play it—pretty impressive.

So here is my question, Harry, to you: is he going to stay? If he’s going to stay for 24 months, because it felt like he’s going to stay, okay? But if he’s really going to stay for 24 months, I’m going Wix, just on financial engineering. It’s nothing against Lovable if he’s going to stay. If he’s going to leave, I’m putting my money in Lovable if I have to pick.

Rory O'Driscoll

Unwavering that Mayol will stay.

Jason Lemkin

Yeah, so I’m not sure I’m going to do it. But if you promise me he’ll stay for 24 months, that arbitrage to that second bucket is so palpable. But he’s got to stay. He’s really good.

I didn’t know him when they bought it. I’ve used Base44, but he’s—

Rory O'Driscoll

If I was on his comp committee, and I’ve done this with some of my existing companies, I would have that guy on an accelerated equity grant program based on upsell of the new AI product across the existing customer base that would make him wildly wealthy.

Double or treble his already hit if he got that penetration up to 20%. And if he's as good as you say, because that's the mission here. If you can get penetration of this product up to 20%, you probably jump 2 buckets, Jason. You probably jump to the $25 billion bucket.

Jason Lemkin

The problem is he could probably raise $1 billion now in a new startup. That's the problem. So you have to compete with that. Now, that's not liquid, right? But if you're running Wix, how do you compete with the fact that, in a sense, he sold really cheap, looking back on it? At the time, in internet time, it seemed like a fair deal, right?

Rory O'Driscoll

It did. The only thing I'll say—and I'm not sharing anything, because I don't know anything—is that I do know from him directly that there is a variable package.

Jason Lemkin

Yeah, fair call.

Rory O'Driscoll

And so he unwaveringly has upside if he hits metrics. If not—

Jason Lemkin

But instead of starting over, in the old days, like 24 months ago, a guy like him would start over, and instead of raising it at a $10 million pre-money valuation, he'd raise it at $50 million pre-money for his next company, or $60 million. Now he can raise it at a $1 billion pre-money valuation for his next startup. I'm not saying it's liquid, but it's a siren call for an aggressive guy. A billion? Yeah, and they'll give you $150 million to start from Andreessen. It's a tough one to say no to if you're ambitious. It's a tough one to say no to: $100 million, $150 million to start, base 45.

Rory O'Driscoll

I just want to say, I don't know a lot about a lot, but I know how to design comp plans. You could give that guy a package at Wix that makes it worth his while. Absolutely.

Jason Lemkin

You can. You can.

Rory O'Driscoll

Because there's only one thing that matters in that damn thing. It's like, get that damn penetration to 20% or 30%, and you have a thing. And if not, you have a 2× revenue-growth, 4× revenue-growth thing. So it's pretty—

Going back to where we started this conversation: mission clarity. I mean, call it war mode, call it what you like. Mission clarity is worth a lot. The mission clarity for that company is: take this existing customer base, who should be using vibe coding to build this stuff. Maybe you have the product, you've bought the product, you have the founder—make it happen.

Jason Lemkin

But here's the question. If they're worth $2 billion today, and if he does $250 million next year, then let's say they're worth $5 billion because of multiple inflation, which it's worth, right? You have to pay him $300 million because he's added $3 billion to your market cap. You have to give him an Elon package to stay. You have to give him 10% of it.

Rory O'Driscoll

I'm not going to devolve down to numbers that I'll get thrown back in my face by every one of my CEOs. And I have to run—

Jason Lemkin

Honestly, I would quit if you didn't give me $300 million.

Rory, we miss you in London, my friend. You need to make it out here next year.

Jason Lemkin

In fact, there's still time. There's still time for next week.

Next week.

Jason Lemkin

There's still time. There's still time.

Rory O'Driscoll

Okay, guys.

Jason Lemkin

We'll fly you out on our nickel.

Rory O'Driscoll

Got it. I will join early next week so we can make sure this stuff works. It'll be crazy.

You're a star, dude.

Rory O'Driscoll

All right. Have fun.

After the pasta. But before we leave you today, now most people who get scammed never talk about it. And if it can happen to tech-savvy professionals, CEOs, and investors, it can happen to anyone. But the problem isn't just losing money, it's that today's scams, they're built differently for a very new world, one where AI can generate convincing messages in seconds and fake sites look more like real sites than the real thing. Traditional tools were not built for this future, and that's why Guardio exists. Guardio is this incredible predictive and proactive engine. It leverages advanced AI threat detection to block highly targeted, socially engineered scams before they ever reach you. From phishing emails and fake login pages to financial fraud, Guardio protects you across the ways people actually live and work online. And security shouldn't be complicated. Guardio continuously monitors across all your accounts and devices, uncovering risks in real time and guiding you to close gaps before attackers exploit them. Trusted by over a million users, Guardio is setting the new standard for personal cybersecurity. Visit guard.io/20vc today to start your seven-day free trial. Because the threats of tomorrow, they're already here, and Guardio is built to stop them. After Guardio protects you from what's out there on the internet, Squarespace helps you put your work out there. Every dream needs a domain. Bring it to life with Squarespace. Your custom domain comes with everything it should. Powerful, built-in security, award-winning support, and tools that help you grow for an upfront, transparent fee. No matter where you are in your journey, do more with a domain. Go to domains.squarespace.com/20vc and get 20% off your dream domain with coupon code VC20. That's VC20. And finally, we have to speak about our newest sponsor. It's Intercom. If you're looking for a way to transform your customer service, let me introduce you to Fin, baby. Fin is the number one AI agent for customer service, resolving up to 93% of customer queries automatically. There is no other agent that can do that. Not 93% of customer queries, okay? No other agent can do that. So why choose Fin? Fin is the best performing AI agent for CS. Fin doesn't just answer questions. It takes actions. It automates the most complex customer queries like refunds, transaction disputes, technical troubleshooting with speed and reliability. I wish my team was speedy and reliable. Beats every competitor in every head-to-head bake-off. Completely configurable and code optional setup. My word. I mean, the benefits just go on and on. It's easy and efficient implementation. It works on any help desk with no tedious migration needs. It's trusted by over 6,000 customer service leaders, including top AI companies like Anthropic, Lovable, Synthesia, Clay, Vanta. So if you're ready to transform your customer service team, scale your support, and give team members time to focus on the really high-level strategic work, learn more about Fin at fin.ai/20vc.